我如何才能从Checkpoint keras模型继续训练？

如何解决我如何才能从Checkpoint keras模型继续训练？

我使用以下代码在2个时期训练第一个检查点模型：

list

要从第一个检查点模型继续训练（第二个检查点模型），请使用以下代码：

def check_units(y_true,y_pred):
if y_pred.shape[1] != 1:
    y_pred = y_pred[:,1:2]
    y_true = y_true[:,1:2]
return y_true,y_pred

def precision(y_true,y_pred):
y_true,y_pred = check_units(y_true,y_pred)
true_positives = K.sum(K.round(K.clip(y_true * y_pred,1)))
predicted_positives = K.sum(K.round(K.clip(y_pred,1)))
precision = true_positives / (predicted_positives + K.epsilon())
return precision

def recall(y_true,1)))
possible_positives = K.sum(K.round(K.clip(y_true,1)))
recall = true_positives / (possible_positives + K.epsilon())
return recall

def fmeasure(y_true,y_pred):
def recall(y_true,y_pred):
    true_positives = K.sum(K.round(K.clip(y_true * y_pred,1)))
    possible_positives = K.sum(K.round(K.clip(y_true,1)))
    recall = true_positives / (possible_positives + K.epsilon())
    return recall

def precision(y_true,1)))
    predicted_positives = K.sum(K.round(K.clip(y_pred,1)))
    precision = true_positives / (predicted_positives + K.epsilon())
    return precision

y_true,y_pred)
precision = precision(y_true,y_pred)
recall = recall(y_true,y_pred)
return 2 * ((precision * recall) / (precision + recall + K.epsilon()))

classifier = Sequential()

classifier.add(Conv2D(6,(3,3),input_shape = (30,30,data_format="channels_last",activation = 'relu'))
classifier.add(MaxPooling2D(pool_size = (2,2)))

classifier.add(Conv2D(6,2)))

classifier.add(Flatten())

classifier.add(Dense(units = 128,activation = 'relu'))
classifier.add(Dense(units = 64,activation = 'relu'))
classifier.add(Dense(units = 1,activation = 'sigmoid'))

opt = Adam(learning_rate = 0.001,beta_1 = 0.9,beta_2 = 0.999,epsilon = 1e-08,decay = 0.0)
classifier.compile(optimizer = opt,loss = 'binary_crossentropy',metrics = ['accuracy',precision,recall,fmeasure])

from keras.preprocessing.image import ImageDataGenerator

train_datagen = ImageDataGenerator(rescale = 1./255,horizontal_flip = True,vertical_flip = True,rotation_range = 180)

validation_datagen = ImageDataGenerator(rescale = 1./255)

training_set = train_datagen.flow_from_directory('/home/dataset/training_set',target_size = (30,30),batch_size = 32,class_mode = 'binary')

validation_set = validation_datagen.flow_from_directory('/home/dataset/validation_set',class_mode = 'binary')

history = classifier.fit_generator(training_set,steps_per_epoch = 208170,epochs = 2,validation_data = validation_set,validation_steps = 89140)

classifier.save('/content/gdrive/My Drive/Checkpoints/Checkpoint_1/Model.h5')

但是我得到此错误的原因是什么？

def check_units(y_true,y_pred)
return 2 * ((precision * recall) / (precision + recall + K.epsilon()))

train_datagen = ImageDataGenerator(rescale = 1./255,class_mode = 'binary')

custom_metrics = {
'precision' : precision,'recall' : recall,'fmeasure' : fmeasure    
}

classifier = load_model('/content/gdrive/My Drive/Checkpoints/Checkpoint_1/Model.h5',custom_objects = custom_metrics)

history = classifier.fit(training_set,validation_steps = 89140)

classifier.save('/content/gdrive/My Drive/Checkpoints/Checkpoint_2/Model.h5')

解决方法

我注意到您每个纪元的步数太大，请尝试使用表达式而不是减小维度的数字输入。

steps_per_epoch = len(input_train)//BATCH_SIZE

考虑到数据集和输入大小，也可以尝试上述操作，这样可以帮助您开始训练。

,

根据我在上面的代码中发现的内容，您可以使用以下代码创建训练和测试数据集。

training_set = train_datagen.flow_from_directory('/home/dataset/training_set',target_size = (30,30),batch_size = 32,class_mode = 'binary')

validation_set = validation_datagen.flow_from_directory('/home/dataset/validation_set',class_mode = 'binary')

如果使用Tensorflow 1.X，一个简单的修复方法类似于Manik所说的。调整拟合函数以采用以下公式int(steps_per_epoch/batch_size)。

history = classifier.fit(training_set,steps_per_epoch = int(208170/batch_size),epochs = 2,validation_data = validation_set,validation_steps = int(89140/batch_size))

如果您使用的是Tensorflow 2.X +，那么以下功能将更适合您的需求。我在此Github问题上找到了答案：https://github.com/tensorflow/tensorflow/issues/25254

#Get your data
training_set...
validation_set...

#Declare the types and shape of your data
types = (tf.float32,tf.int32)
shapes = ((512,512,3),(2,))
ds_train = tf.data.Dataset.from_generator(lambda: training_set,types,shapes).shuffle(steps_per_epoch).batch(batch_size)
ds_test = tf.data.Dataset.from_generator(lambda: validation_set,shapes).shuffle(steps_per_epoch).batch(batch_size)

# usage in model
model.fit(training_set,validation_data=validation_set,epochs=10,verbose=True,callbacks=[visualize,tensorboard])